Hoid was founded by three engineers from Databricks, Tenstorrent and Cambridge, backed by a16z Speedrun. 

At Hoid, we are building an agentic AI compiler that turns AI models into highly optimized implementations for any hardware.

Instead of relying on fixed compiler rules or hand-tuned kernels, Hoid gives the compiler an agentic layer that can inspect a full model, find bottlenecks, test alternatives, and discover optimizations specific to each model and chip.

This makes a new kind of AI deployment possible: models that automatically adapt to the hardware they run on, pushing toward peak performance across every environment.